Re: [LinuxBIOS] #57: libusb host program for PLX NET20DC debug device

[ebiederm@xmission.com (Eric W. Biederman)]

Peter Stuge <stuge-linuxbios@cdy.org> writes:

> On Fri, Dec 01, 2006 at 11:19:16AM -0800, Greg KH wrote:
>> Well, earlyprintk will not work, as you need PCI up and running.
>
> Not all of it though. LinuxBIOS will probably do just enough PCI
> setup to talk to the EHCI controller and use the debug port _very_
> soon after power on.

Right.  For LinuxBIOS not a problem for earlyprintk in the kernel
somethings might need to be refactored.  The challenge in the kernel
is we don't know at build to how to do a pci_read_config...

The other hard part early in the kernel is the fact that the
bar is memory mapped I/O.  Which means it will need to get mapped
into the kernels page tables.

>> And I have some code that barely works for this already, perhaps
>> Eric and I should work together on this :)
>
> I would be interested in having a look at any code for it too.

Sure, I will send it out shortly.  I currently have a working
user space libusb thing (easy, but useful for my debug) and 
a rude read/write to the bar from user space program that
allowed me to debug the worst of the state machine from user
space.  I don't think I have the state setup logic correct yet
but that is minor in comparison.

I really wish the EHCI spec had made that stupid interface 16 bytes
instead of 8 or had a way to chain multiple access together.  The
we could have used a normal usb cable.  As it is most descriptors
are 1 byte to big to read.

>> Yes, that will work just fine today using the usb-serial generic
>> driver.
>
> Ugh. I did not know it was that generic. The irony is that I always
> ask other libusb users to check the kernel drivers to see if they
> really need to write a libusb app.
>
>
>> I'll knock up a "real" driver for the device later today and send
>> it to Linus, as it's trivial to do so, and will make it simpler
>> than using the module parameters.
>
> Awesome. Thanks!

Yep. It looks like it sufficient generic.  The Maximum packet size
appears to be reported correctly for writes which is the tidbit
I was worried about but otherwise it is just a pair of bulk transfer
endpoints.

Eric